Why Machine Learning Made OSINT Practical for Supply Chains

For decades, the frustrating truth about supply chain risk was that the warning signs were usually public — and usually missed. The sanction, the lawsuit, the local-language news report, the filing: all sitting in the open, and all effectively invisible because no team could read a fraction of it in time. Open-source intelligence, the discipline of turning public data into insight, held the answers. What it lacked was a way to process them at the scale and speed a real supplier base demands.

That is the gap the founder's research at Copenhagen Business School set out to close, and it is the reason OSINT for supply chain risk only became practical recently: not because the data appeared, but because machine learning finally made it readable.

Two facts frame the challenge. First, the sheer quantity of relevant public information is far beyond human reading capacity: news in dozens of languages, millions of company records, filings, registries, and sanctions updates, changing every day. Second, most of it does not even surface in a normal search — by industry estimates the great majority of the web is not indexed by mainstream engines like Google or Bing, sitting behind logins, paywalls, databases, and sheer overload.

Put together, those two facts kill the naive approach. "Have someone Google the supplier" fails because the searcher can only read a sliver, in one or two languages, of the small fraction of the web that is searchable at all. The intelligence was public but not accessible — a distinction that mattered enormously.

Machine learning is what converts that firehose into signal. Natural-language processing reads and translates text across languages and pulls out entities and events — this company, this location, this kind of risk. Text-mining techniques sift enormous volumes to separate the relevant few items from the irrelevant many. Neural-network models classify and score, learning to tell a genuine risk event from background noise. None of these replace judgement; together they do the reading, filtering, and first-pass classification that no human team could.

The critical output is resolution: matching a story to the specific supplier it concerns, rather than to a keyword or a country. That entity-resolution step — done across a vast universe of companies — is what turns "a lot of news" into "a risk alert about your tier-2 supplier in Izmir." It is the difference between a clipping service and a risk system.

The founder's research did not stop at theory; it built and tested a working prototype that demonstrated OSINT data capture across roughly 490 million companies — evidence that machine-learning-enhanced OSINT could operate at genuine supply-chain scale rather than on a handful of hand-picked suppliers. Intellens is that research, productised: the same principles — read broadly, translate, resolve to the supplier, score, alert — turned into a platform procurement teams can use.

The takeaway for a buyer is simple. The reason continuous, comprehensive supplier monitoring is possible now, and was not a decade ago, is machine learning applied to open sources. To see it working on your own network, request a demo, or read more in the Risk Center.